Journal: bioRxiv

Article Title: The multiscale distribution of radiation-induced DNA damage and its impact on local genome structure

doi: 10.1101/2025.08.25.672161

Figure Lengend Snippet: a. Scatter plots showing the relationship between the length of chromosomes and the number of robust breaks in BJ-5ta (top) and GM12878 (bottom) cells. b. Scatter plots exhibiting the relationship between the position of chromosomes within the nucleus and the number of robust breaks in BJ-5ta (top) and GM12878 (bottom) cells. All p-values are from Spearman correlation test as indicated in figures.

Article Snippet: BJ-5ta (RRID: CVCL_6573) and GM12878 (RRID: CVCL_7526) cells were purchased from ATCC (CRL-4001) and Coriell respectively.

Techniques:

Journal: bioRxiv

Article Title: The multiscale distribution of radiation-induced DNA damage and its impact on local genome structure

doi: 10.1101/2025.08.25.672161

Figure Lengend Snippet: a. The distribution of robust breaks in chromosome X after IR BJ-5ta (top) and GM12878 (bottom) cells. Each tick mark represents a 50 kb bin classified as a robust break superimposed on the chrX diagram. b. Boxplots representing the break frequency in 50 kb windows on the p and q arm of chromosome X in GM12878 cells. c. Histogram showing the sum of RNA-seq signal normalized by chromosome arm length for the p and q arm of chromosome X in GM12878 cells.

Article Snippet: BJ-5ta (RRID: CVCL_6573) and GM12878 (RRID: CVCL_7526) cells were purchased from ATCC (CRL-4001) and Coriell respectively.

Techniques: RNA Sequencing

Journal: bioRxiv

Article Title: The multiscale distribution of radiation-induced DNA damage and its impact on local genome structure

doi: 10.1101/2025.08.25.672161

Figure Lengend Snippet: a, b. Boxplots representing the break frequency in each 50 kb window for all chromosomes in replicate 1 in BJ-5ta ( a ) and GM12878 ( b ) cells. Boxes are colored by chromosomes. Dashed lines represent the median value across all chromosomes. Boxes show 25 th percentile, median, and 75 th percentile with whiskers showing up to 1.5 times the interquartile range. Numbers of bins on each chromosome are shown above each box.

Article Snippet: BJ-5ta (RRID: CVCL_6573) and GM12878 (RRID: CVCL_7526) cells were purchased from ATCC (CRL-4001) and Coriell respectively.

Techniques:

Journal: bioRxiv

Article Title: The multiscale distribution of radiation-induced DNA damage and its impact on local genome structure

doi: 10.1101/2025.08.25.672161

Figure Lengend Snippet: a. Boxplots representing the break frequency in each 50 kb window for A (red) and B (blue) compartments in BJ-5ta (left) and GM12878 (right) cells in replicate 1. b. Bar plots showing the counts of 250 kb bins with (sky blue) or without (rose) robust breaks within compartments that shifted between control and 30 min post-irradiation, including A to B and B to A, in BJ-5ta (left) and GM12878 (right) cells. c. Boxplots representing the counts of robust breaks per megabase within TAD boundary regions (dark purple) or non-boundary regions (lime) in BJ-5ta (left) and GM12878 (right) cells. Boxes show 25 th percentile, median, and 75 th percentile with whiskers showing up to 1.5 times the interquartile range. Numbers of bins on each chromosome are shown above each box. P-value calculated using Mann-Whitney U test.

Article Snippet: BJ-5ta (RRID: CVCL_6573) and GM12878 (RRID: CVCL_7526) cells were purchased from ATCC (CRL-4001) and Coriell respectively.

Techniques: Control, Irradiation, MANN-WHITNEY

Journal: bioRxiv

Article Title: The multiscale distribution of radiation-induced DNA damage and its impact on local genome structure

doi: 10.1101/2025.08.25.672161

Figure Lengend Snippet: d. Hi-C contact matrix centered on a selected robust DSB in nonirradiated (control) BJ-5ta ( a ) and GM12878 ( d ) (50-kb resolution, 5-Mb window). b, e. Hi-C contact difference matrix (30 min after X-ray – Control) centered on the robust DSB in BJ-5ta ( b ) and GM12878 ( e ) (50-kb resolution, 5-Mb window). c, f. Hi-C contact difference matrix (24 h after X-ray - Control) centered on the robust DSB in BJ-5ta ( c ) and GM12878 ( f ) (50-kb resolution, 5-Mb window).

Article Snippet: BJ-5ta (RRID: CVCL_6573) and GM12878 (RRID: CVCL_7526) cells were purchased from ATCC (CRL-4001) and Coriell respectively.

Techniques: Hi-C, Control

Journal: bioRxiv

Article Title: Expanding the DNA Motif Lexicon of the Transcriptional Regulatory Code

doi: 10.1101/2025.07.09.662874

Figure Lengend Snippet: (A) MPRA library design. The MPRA library comprises 30 bp oligonucleotides containing either simple TF motifs (n=153, with 20 instances of each) or CEs (n=3,795, with 4 instances of each). For simple motifs, each wildtype (WT) sequence is paired with a mutated version (M). For cCEs, each wildtype sequence is paired with three mutated variants: left motif mutated (M1), right motif mutated (M2), and both motifs mutated (M1, M2). Mutations were introduced at two consecutive positions with the highest information content in each TF binding site. Specifically, A was changed to C, T to G, and vice versa. The complexity of the library was n=60,714 sequences, and the activity of each sequence was assayable with >100 linked barcodes (see Methods for additional details). (B) MPRA analysis pipeline. The analysis of the TF-motif MPRA datasets is performed at three levels: barcode, sequence, and motif. In the barcode-level analysis, anomalous barcodes with excessively high read counts are removed. The remaining barcodes are then aggregated to their respective TF motif sequences, enabling quantification of sequence-level activity using DESeq2 to compare total barcode counts in RNA vs. DNA. At the motif level, simple motif activities are assessed by comparing the RNA/DNA log2 fold change (log2FC) of the wildtype motif-containing oligonucleotides with the log2FC of matched mutated control oligonucleotides. The activities of composite motifs are similarly estimated by comparing the log2FC values of the RNA/DNA ratios of sequences containing the wild type cCE with those of their double-mutant counterparts (see Methods ). (C) MPRA assays of simple and cCE motif activities in the indicated human cell lines. The MPRA library was transfected into K562 erythro-myeloid cells (6 replicates), GM12878 B cells (5 replicates), and Jurkat T cells (5 replicates) (see Methods ). Box plots showing the distributions of the transcriptional activities of simple motifs and cCEs in the three cell lines. Activating and repressive motifs are distinguished by red and blue color codes, respectively. (D) Counts of activating motifs. Bar plots depict the number of activating simple motifs and cCEs in K562, GM12878, and Jurkat cells. The shaded bars highlight activating cCEs that are composed of simple motifs that lack detectable transcriptional activity on their own in the indicated cellular context. (E) Shared and cell type-specific activating cCEs. Pie chart illustrating the numbers of shared and cell-type specific activating cCEs among the K562, GM12878, and Jurkat cell lines. (F–H) Examples of activating cCEs. Sequence logos (PWMs) represent cCEs, while bar plots show the transcriptional activities of the paired wild-type oligonucleotides and their mutated variants. The dashed lines relate activity of each wildtype oligonucleotide with its mutant counterparts. (I–K) Synergistic and antagonistic interactions within TF motifs comprising cCEs. Volcano plots highlighting cCEs that exhibit significant synergy or antagonism between their two constituent simple TF motifs in the three cell lines.

Article Snippet: GM12878 cells (Coriell Institute for Medical Research) were cultured in RPMI medium (Corning, 10-040-CV) supplemented with 15% FBS (Gibco A52568-01), 1% GlutaMAX (Gibco 35050-061) and 1% Pen-Strep (Corning, 30-002-CI).

Techniques: Sequencing, Binding Assay, Activity Assay, Control, Mutagenesis, Transfection

Journal: bioRxiv

Article Title: Expanding the DNA Motif Lexicon of the Transcriptional Regulatory Code

doi: 10.1101/2025.07.09.662874

Figure Lengend Snippet: (A) Barcode count distribution. Histograms showing the distribution of unique barcode counts recovered from RNA for TF motif library sequences after transfection in K562 (left), GM12878 (middle), and Jurkat cells (right). Sequences with ≥30 unique barcodes (dashed line) were used for downstream analyses. (B) Reproducibility of the biological replicates. Heatmap displaying pairwise Pearson correlation coefficients for aggregated barcode counts linked with a given TF motif sequence across RNA and DNA biological replicates. (C) Schematic illustrating the determination of simple and composite motif activities from the MPRA datasets. For simple motifs, motif activity was calculated using DEseq2 by comparing the RNA/DNA log2 fold change (log2FC) of motif-containing sequences (sequence activity) as test values and the RNA/DNA log2 fold change (log2FC) of matched mutated control sequences as reference values. For composite motifs, activities were similarly assessed by using the RNA/DNA log2FC of CE motif-containing sequences with those of matched double-mutant controls. To determine individual motif contributions within a CE, the RNA/DNA log2FC values of sequences carrying single-motif mutations were analyzed, with those from matched double-mutant sequences used as controls. (D) Numbers of functional simple and composite TF motifs revealed by MPRA assays. Table summarizing the counts of functional (activating or repressing) simple and composite motifs identified in K562, GM12878, and Jurkat cells, along with their median sequence and motif activities. (E) Comparisons of cCE activities across cell types. Scatter plots of cCE motif activity (log2FC) measured in GM12878 vs. K562 (left), GM12878 vs. Jurkat (middle), and Jurkat vs. K562 (right) cells. The green and blue dots indicate cell type-specific activating or repressing cCEs in the indicated cell types. The red dots are cCEs that are activating or repressing in both cell types. The gray dots are cCEs with no detectable activity in the indicated cell types. Representative simple TF motifs that are constituents of cell-type specific CEs are shown in the left panel.

Article Snippet: GM12878 cells (Coriell Institute for Medical Research) were cultured in RPMI medium (Corning, 10-040-CV) supplemented with 15% FBS (Gibco A52568-01), 1% GlutaMAX (Gibco 35050-061) and 1% Pen-Strep (Corning, 30-002-CI).

Techniques: Transfection, Sequencing, Activity Assay, Control, Mutagenesis, Functional Assay

Journal: bioRxiv

Article Title: Expanding the DNA Motif Lexicon of the Transcriptional Regulatory Code

doi: 10.1101/2025.07.09.662874

Figure Lengend Snippet: (A–C) Representative CEs demonstrating enrichment in co-bound genomic regions of cognate TFs. The top panels display sequence logos of the CEs (PWMs), whereas the bar plots show the activities of the wild-type oligonucleotides and their mutated variants (connected by dashed lines) in the indicated cells. For each CE, the simple TF motif was used to query CistromeDB for a cognate TF ChIP-seq dataset. Venn diagrams showing the numbers of co-bound versus singly bound regions for the relevant pairs of TFs. Heatmaps depict the enrichment of CE configurations (n=64) in co-bound TF ChIP-seq peak regions relative to singly bound peak regions. The CE configurations assayed in the MPRA library are highlighted with green boxes in the heatmaps and correspond to the PWMs displayed in the top panels. (D, E) Diversity of CEs comprising AP-1 or KLF motifs. Sequence logos of the AP-1 ( D ) or KLF ( E ) families of CEs are shown on the left. The bar plots show the measured sequence, motif, and synergistic activities of each CE in K562, GM12878, and Jurkat cells. These CEs were enriched in corresponding TF ChIP-seq datasets (see panels A-C and ).

Article Snippet: GM12878 cells (Coriell Institute for Medical Research) were cultured in RPMI medium (Corning, 10-040-CV) supplemented with 15% FBS (Gibco A52568-01), 1% GlutaMAX (Gibco 35050-061) and 1% Pen-Strep (Corning, 30-002-CI).

Techniques: Sequencing, ChIP-sequencing

Journal: bioRxiv

Article Title: Expanding the DNA Motif Lexicon of the Transcriptional Regulatory Code

doi: 10.1101/2025.07.09.662874

Figure Lengend Snippet: (A) Model overview. GRACE is a neural network model that integrates convolutional layers to learn TF binding motifs and employs self-attention with positional encoding to capture interactions among these motifs. See Methods for details. (B) Prediction accuracy of GRACE. Scatterplot comparing GRACE-predicted log2FC transcriptional activities for the withheld set of sequences (n=3,900) from the training set (n=49,042) with their experimentally measured MPRA activities in GM12878 cells. The Pearson correlation coefficient is indicated. (C) Correlations of the GRACE predictions with the MPRA replicates. Heatmap showing pairwise correlations between GRACE predictions and MPRA measurements across individual biological replicates, as well as with the aggregated set of MPRAs in GM12878 cells. (D) Generating transcriptional activity contribution scores with GRACE. Illustration of GRACE assigned activity contribution scores to each base pair using in silico saturation mutagenesis. The indicated 30 bp oligonucleotide, corresponding to the displayed CE position weight matrix (PWM), was systematically mutated to evaluate the impact of individual base pairs on its transcriptional activity using GRACE model trained with the GM12878 MPRA dataset (heatmap). The resulting contribution weight matrix (CWM) is displayed below the heatmap. (E) Histogram showing the false discovery rate (FDR) of the similarity between CEseek PWMs and GRACE CWMs for activating cCEs. All the activating cCE sequences (30 bp) in the MPRA library were used to generate CWMs via the GRACE model of GM12878 cells, which were then compared to their corresponding PWMs. The histogram shows the distribution of median similarity false discovery rate (FDR) values between the PWMs and CWMs, with the fraction of matching cCEs (FDR < 0.05) indicated in the panel. (F) Examples of matched CEseek PWMs with GRACE predicted CWMs. The FDR values for each pair are indicated.

Article Snippet: GM12878 cells (Coriell Institute for Medical Research) were cultured in RPMI medium (Corning, 10-040-CV) supplemented with 15% FBS (Gibco A52568-01), 1% GlutaMAX (Gibco 35050-061) and 1% Pen-Strep (Corning, 30-002-CI).

Techniques: Binding Assay, Activity Assay, In Silico, Mutagenesis

Journal: bioRxiv

Article Title: Expanding the DNA Motif Lexicon of the Transcriptional Regulatory Code

doi: 10.1101/2025.07.09.662874

Figure Lengend Snippet: (A) Ranking of the OCRs according to their GRACE predicted transcriptional activities. Scatter plot displaying OCRs in GM12878 cells ranked by their GRACE-predicted transcriptional activities (see Methods ). (B) Box plots illustrating the distributions of H3K27ac ChIP-seq signals (reads per million, RPMs) across the GRACE-ranked OCRs in GM12878 cells, as ordered in panel (A) . See Table S9 for the H3K27Ac dataset. (C, D) GRACE predicted transcriptional activities in the promoter and enhancer regions of GM12878 cells. Promoters were defined by having an annotated TSS and an activity measured by GRO-cap in GM12878 cells of >1 TPM (see Table S9 for the dataset). Promoter regions were aligned based on annotated TSSs ( C ). Enhancers lacked an annotated TSS within a 1000 bp and were aligned based on the center of one of the GRACE-predicted transcriptionally active sequences (seqlets) ( D ) (see Methods ). The aligned promoter or enhancer sequences were clustered using K-means, and the heatmaps display the median GRACE-predicted transcriptional activities. The ridge plots above and below each heatmap represent the median GRO-cap signals generated from the promoter sequences in each cluster on the sense/forward and antisense/reverse strands, respectively. The numbers (#) of aligned promoter or enhancer sequences in each K-means cluster are indicated on the right of each panel. (E, F) Analysis of TF motif contributions to chromatin accessibility and transcriptional activity using ChromBPNet and GRACE models. Scatter plots depict the contributions of simple (gray) and CE (red) motifs to chromatin accessibility (x-axis, derived using ChromBPNet) and transcriptional activity (y-axis, derived using GRACE) in K562 ( E ) and GM12878 ( F ) cells. Motif contributions to chromatin accessibility or transcriptional activity are defined as fold changes between contribution scores of motif sequences (ChromBPNet or GRACE) compared with the contribution scores of similarly modeled 50 bp flanking regions (see Methods ). Representative simple TF motifs that are contained either within highly active CEs or within CEs that differentially contribute to chromatin accessibility and transcriptional activity in indicated cell contexts are displayed on plots. (G–I) Examples of activating and repressing CEs with their motif contributions to chromatin accessibility and transcriptional activity. The left and right columns in each panel illustrate the simple TF motifs that comprise the corresponding CEs in the middle column. The first row in each panel displays sequence logos (PWMs) based on CEseek. The second and third rows provide the aggregated contribution weight matrices (CWMs) for chromatin accessibility (second row) and transcriptional activity (third row) using ChromBPNet and GRACE models, respectively. The dashed gray lines indicate baseline contributions estimated for 50 bp flanking regions (see Methods ).

Article Snippet: GM12878 cells (Coriell Institute for Medical Research) were cultured in RPMI medium (Corning, 10-040-CV) supplemented with 15% FBS (Gibco A52568-01), 1% GlutaMAX (Gibco 35050-061) and 1% Pen-Strep (Corning, 30-002-CI).

Techniques: ChIP-sequencing, Activity Assay, Generated, Derivative Assay, Sequencing

Journal: bioRxiv

Article Title: Expanding the DNA Motif Lexicon of the Transcriptional Regulatory Code

doi: 10.1101/2025.07.09.662874

Figure Lengend Snippet: (A, B) GRACE predicted transcriptional activities in the promoter and enhancer regions of GM12878 cells. These panels show the remaining promoter and enhancer K-means clusters from the analysis detailed in , . (C) Box plots illustrating the distribution of ATAC-seq signals (RPMs) across the GRACE-ranked open chromatin regions displayed in . (D) Concordance of GRACE and ChromBPNet predictions at TSSs. Plots show the mean ChromBPNet and GRACE contribution scores ±500 bp of transcription start sites (TSSs) of active promoters in GM12878 cells. Profiles are compared by normalizing to maximum contribution scores for transcriptional activity or chromatin accessibility. (E–G) Examples of activating simple motifs or CEs with their chromatin accessibility and transcriptional activity contributions (See for details).

Article Snippet: GM12878 cells (Coriell Institute for Medical Research) were cultured in RPMI medium (Corning, 10-040-CV) supplemented with 15% FBS (Gibco A52568-01), 1% GlutaMAX (Gibco 35050-061) and 1% Pen-Strep (Corning, 30-002-CI).

Techniques: Activity Assay

Journal: bioRxiv

Article Title: Expanding the DNA Motif Lexicon of the Transcriptional Regulatory Code

doi: 10.1101/2025.07.09.662874

Figure Lengend Snippet: (A) Identification of a CREB::ETS CE using CEseek and ChIP-seq data for CREB1 and SPI1. The Venn diagram showing the overlap of the CREB1 and SPI1 binding regions (ChIP-seq) in GM12878 cells. Co-binding regions were further analyzed with CEseek, using singly bound regions as controls, to identify statistically enriched CREB::ETS configurations (heatmap). One such configuration corresponding to the motif impacted by the GWAS allele rs6011530 is highlighted with an orange box. The corresponding CREB::ETS CE PWM is shown below the heatmap. (B-E) Examples of autoimmune disease-associated GWAS variants with their GRACE interpretations. Sequence logos showing GRACE nucleotide contributions for the reference and alternative alleles along with their annotated simple motifs ( B, C ) or CEs ( D, E ). All alternative alleles are predicted to result in gain-of-function. The heatmaps in the middle row depict in silico saturation mutagenesis predictions generated by GRACE for the reference sequences.

Article Snippet: GM12878 cells (Coriell Institute for Medical Research) were cultured in RPMI medium (Corning, 10-040-CV) supplemented with 15% FBS (Gibco A52568-01), 1% GlutaMAX (Gibco 35050-061) and 1% Pen-Strep (Corning, 30-002-CI).

Techniques: ChIP-sequencing, Binding Assay, Sequencing, In Silico, Mutagenesis, Generated

Journal: Nature Communications

Article Title: Unveiling chromatin dynamics with virtual epigenome

doi: 10.1038/s41467-025-58481-3

Figure Lengend Snippet: a Predicted Hi-C matrices from four methods in four percentile ranges: The visualization compares the ground truth Hi-C interaction matrices with those imputed by EpiVerse, Hi-C-Reg, C.Origami, and Orca across various percentile ranks (PR99, PR75, PR50, PR25). EpiVerse’s predictions are shown to closely approximate the ground truth, particularly in the top 0–25% range, indicating high-quality imputation of chromatin interactions. b Performance metrics comparison: The violin plots depict the distribution of Pearson correlation coefficient, Spearman correlation coefficient, SSIM, and O/E correlation values for each method ( n = 5243), with EpiVerse outperforming other models across all metrics, signifying its superior predictive accuracy. Violin plot elements are defined as follows: center point represents the median; box limits indicate the upper and lower quartiles; whiskers extend to 1.5 times the interquartile range from the quartiles; and the overall shape represents the kernel density estimation of the data distribution. c Distance-stratified correlation comparison: This figure displays Pearson correlation coefficient values stratified by genomic distance, illustrating EpiVerse’s enhanced performance in predicting long-range interactions, particularly evident at distances beyond 0.5 Mb. d Cross-cell type prediction validation: The table shows the SSIM performance for EpiVerse, Hi-C-Reg, C.Origami, and Orca across different cell types (IMR90, GM12878, K562), with EpiVerse consistently delivering high-quality predictions, demonstrating robust cross-cell-type generalization.

Article Snippet: The ChromHMM data for IMR90 (E017), GM12878 (E116), and K562 (E123), based on the 25-state, 12-mark chromatin state model from the ROADMAP Epigenomics project, were also used for training models and are available at the ROADMAP Epigenomics portal https://egg2.wustl.edu/roadmap/web_portal/index.html .

Techniques: Hi-C, Comparison, Biomarker Discovery